Parsimonious and Profligate Approaches to the 
Question of Discourse Structure Relations 
Eduard H. Hovy* 
Information Sciences Institute of USC 
4676 Admiralty Way 
Marina del Rey, CA 90292-6695 
Telephone: 213-822-1511 
Email: HOVY~ISI.EDU 
Abstract 
To computationalists investigating the structure of co- 
herent discourse, the following questions have become 
increasingly important over the past few years: Can one 
describe the structure of discourse using interclausal 
relations? If so, what interclausal relations are there? 
How many are required? A fair amount of controversy 
exists, ranging from the parsimonious position (that 
two intentional relations suffice) to the profligate posi- 
tion (that an open-ended set of semantic/rhetorical re- 
lations is required). This paper outlines the arguments 
and then summarizes a survey of the conclusions of 
approximately 25 researchers -- from linguists to com- 
putational linguists to philosophers to Artificial Intelli- 
gence workers. It classifies the more than 350 relations 
they have proposed into a hierarchy of increasingly se- 
mantic relations, and argues that though the hierarchy 
is open-ended in one dimension, it is bounded in the 
other and therefore does not give rise to anarchy. Evi- 
dence for the hierarchy is mentioned, and its relations 
(which are rhetorical and semantic in nature) are shown 
to be complementary to the two intentional relations 
proposed by the parsimonious position. 
How Many Interclausal Dis- 
course Coherence Relations? 
This paper proposes an answer to an issue that 
keeps surfacing in the computational study of the 
nature of multisentential discourse. 
It has been argued fairly generally that mul- 
tisentence texts (specifically, short texts such 
*This work was supported by the Rome Air Development 
Center under RADC contract FQ7619-89-03326-0001. 
as paragraphs) are coherent by virtue of the 
rhetorical or semantic relationships that hold 
among individual clauses or groups of clauses (see, 
for example, \[Aristotle 54, Hobbs 79, Grimes 75, 
Mann & Thompson 88\]. In this view, a text is only 
coherent when the speaker aids the heater's infer- 
ential understanding processes by providing clues, 
during the discourse, as to how the pieces of the 
text interrelate. Such clues are often cue words and 
phrases such as "in order to" (signalling a purpose 
for an action) or "then" (signalling the next entity 
in some temporal or spatial sequence); but they can 
also be shifts in tense and mode (such as in "She 
was gone. Had she been there, all would have been 
well"), and even appropriate pronominalizations. 
Various researchers in various intellectual sub- 
fields have produced lists of such relations for 
English. Typically, their lists contain between 
seven and thirty relations, though the more de- 
tailed the work (which frequently means the closer 
the work is to actual computer implementation), 
the more relations tend to be named. I have 
collected the lists of over 25 researchers -- from 
philosophers (e.g., \[Toulmin 58\]) to linguists (e.g., 
\[Quirk & Greenbaum 73, Halliday 85\]) to compu- 
tational linguists (e.g., \[Mann & Thompson 88, 
Hobbs 79\]) to Artificial Intelligence researchers 
(e.g., \[Schank & Abelson 77, Moore & Paris 89, 
Dahlgren 88\]) -- amounting to a total of more than 
350 relations. The researchers and their lists ap- 
pear below. 
In this paper, I will call the assumption of these 
researchers, namely that some tens of interclausal 
relations are required to describe the structure of 
English discourse, the Profligate Position. 
128 
Unfortunately, the matter of interclausal rela- 
tions is not simple, and not everyone agrees with 
this position. These relations are seldom explicitly 
signalled in the text, and even when they are, they 
seem to take various forms particular to their use 
in context. This fact has led some researchers, no- 
tably \[Grosz & Sidner 86\], to question the wisdom 
of identifying a specific set of such relations. They 
argue that trying to identify the "correct" set is a 
doomed enterprise, because there is no closed set; 
the closer you examine interclausal relationships, 
the more variability you encounter, until you find 
yourself on the slippery slope toward the full com- 
plexity of semantics proper. Thus though they do 
not disagree with the idea of relationships between 
adjacent clauses and blocks of clauses to provide 
meaning and to enforce coherence, they object to 
the notion that some small set of interclausal rela- 
tions can describe English discourse adequately. 
As a counterproposal, Grosz and Sidner sidestep 
the issue of the structure of discourse im- 
posed by semantics and define two very ba- 
sic relations, DOMINANCE and SATISFACTION- 
PRECEDENCE, which carry purely intentional (that 
is, goal-oriented, plan-based) import. They use 
these relations in their theory of the structure of 
discourse, according to which some pieces of the 
text are either subordinate to or on the same level 
as other pieces, with respect to the interlocutors' 
intentions. I will call this position, namely that two 
interclausal relations suffice to represent discourse 
structure, the Parsimonious Position. 
From the point of view of text analysis, the 
Parsimonious approach seems satisfactory. Cer- 
tainly one can analyze discourse using the two in- 
tentional relations. However, from the point of 
view of text generation, this approach is not suf- 
ficient. Practical experience has shown that text 
planners cannot get by on intentional considera- 
tions alone, but need considerably more rhetor- 
ical and semantic information in order to con- 
struct coherent text (there are many examples; see 
\[McKeown 85, Hovy 88a, Moore & Swartout 88, 
Paris 88, Rankin 89, Cawsey 89\]). In practical 
terms, this means that text planning systems re- 
quire a rich library of interclausal relations. 
Questions such as 
• Does one really need semantic and/or rhetori- 
cal discourse structure relations? 
• Just how many such relations are there? 
• What is their nature? How do they relate to 
the two intentional relations? 
will not go away. Until it is resolved to the satisfac- 
tion of the adherents both positions, further work 
on text planning and discourse analysis is liable to 
continue getting stranded on the rocks of misunder- 
standing and disagreement. This paper suggests a 
compromise that hopefully opens up the way for 
further development. 
An Unsatisfactory Solution 
How can one reconcile the two sides? That is to 
say, how can one build a library of interclausal re- 
lations that are simultaneously expressive enough 
to satisfy the requirements of text planning systems 
but do not simply form an unbounded ad hoc col- 
lection of semantic relations with no regard to the 
intentional ones? 
One answer is to produce a two-dimensional or- 
ganization of relations, with one dimension con- 
strained in the number of relations and the other 
unconstrained (and increasingly semantic in na- 
ture; see Objection 1 below). Such organization is 
a hierarchic taxonomy of limited width but of un- 
bounded depth; the more general a relation is, the 
higher it is in the hierarchy and the fewer siblings 
it has. 
An appealing hierarchy is shown in Figure 1. It 
displays a number of very desirable features. In 
particular, the top few levels are strictly bounded: 
no logical alternatives to ASYMMETRIC and SYM- 
METRIC exist, and one level lower, under ASYM- 
METRIC, following Grosz and Sidner there is no 
need to use any other relation than DOMINATES 
and SATISFACTIONPRECEDES at that level. In- 
creasingly detailed relations appear at lower lev- 
els, which (as is discussed below) remain relatively 
bounded. Still, the more one specifies a particu- 
lar relation to distinguish it from others, the more 
semantic it necessarily becomes (since increasing 
specification invariably introduces additional se- 
mantic features; that is the nature of the specializa- 
tion process), and the lower it appears in the hierar- 
chy. Though one does eventually approach the full 
complexity of semantics proper, the approach is not 
unprincipled; each relation is always constrained by 
its position in the hierarchy and inherits much of 
its structural and other features from its ancestors. 
In this scheme, one can (and the Parsimo- 
nious do) perform discourse analysis and study dis- 
course structure wholly at the level of DOMINATES 
129 
Figure 1: An Unsatisfactory Attempt at Hierarchicalizing Interclausal Relations. 
I 
I I Asymmetric Symmetric 
I I 
I I I I 
Dominates SatisfactionPrecedes Comparative LoEicalRelation 
I I I I 
........................ 
- Circumstance - Sequence - Contrast - Conjunction 
- Elaboration - Restatement - Comparison - Disjunction 
- etc. - etc. - etc. - etc. 
and SATISFACTIONPRECEDES, and never come into 
conflict with the structural descriptions found em- 
pirically by the Profligate. One is simply not being 
as specific about the particular interclausal rela- 
tions that make up the discourse. 
However, this taxonomy is unsatisfactory. It is 
impossible in practise to place into the hierarchy 
with certainty most of the relations found necessary 
by the Profligate. For example, the relation CAusE 
(of various kinds) is one of the most agreed-upon re- 
lations. But is it to be classified as a type of DOM- 
INATES or of SATISFACTIONPRECEDES? Though it 
seems able to function either way, this question is 
impossible to answer, since none of the concepts 
involved are clearly enough defined (certainly no- 
body has provided a general definition of CAUSE -- 
how could one?; it has been the topic of centuries 
of philosophical debate. And even the limited def- 
inition required for the purposes of Computational 
Linguistics in a particular application domain with 
a given ontology of terms has not been provided 
satisfactorily yet). 
A Better Solution 
The answer to the dilemma: It is a mistake to clas- 
sify rhetorical and semantic relations under the re- 
lations DOMINATES and SATISFACTIONPRECEDES. 
This insight does not destroy the hierarchy; most 
of its desirable properties are maintained. It does 
mean, however, that a new top-level organization 
must be found and that the role of intentional re- 
lations vis £ vis rhetorical/semantic relations must 
be explained. I address the first point here, and 
the role of the intentional relations later on. 
I have not found any highly compelling top-level 
organization. Ideally, the top level should parti- 
tion the relations into a few (say, three or four) 
major groups that share some rhetorical or se- 
mantic property. In the absence of a more com- 
pelling suggestion (for which I continue to search), 
I use here the top-level trifurcation of \[Halliday 85\], 
which is based, roughly speaking, on the "semantic 
distance" between the contents of the two clauses. 
The three relations are ELABORATION, ENHANCE- 
MENT, and EXTENSION. ELABORATION relations 
hold between entities and their immediate con- 
stituents or properties, and have a definitional fla- 
vor; ENHANCEMENT relations hold between enti- 
ties and their circumstances of place, time, manner, 
etc.; and EXTENSION relations hold between enti- 
ties and more distant entities such as causes, follow- 
ups, contrasts, etc. Halliday's classification has 
been modified and regularized by Matthiessen (per- 
sonM communication). For want of compelling ar- 
guments to the contrary, I use Matthiessen's mod- 
ification of I-Ialliday's ideas as the basis for the top 
level of the hierarchy. 
In order to construct the hierarchy, I collected 
over 350 relations from 26 researchers in various 
fields. I merged the relations, coming up with a 
set of 16 on the second tier of the hierarchy, and 
then classified more particular subrelations where 
appropriate. The hierarchy of interclausal rela- 
tions is given in Figure 2. The number associated 
with each relation indicates the number of different 
researchers who have listed the relation and may 
serve as a vote of confidence in it. 
The following list contains each relation in the 
hierarchy together with its proposers (identified 
130 
Elaboration (101 
Figure 2: A Hierarchy of InterclausalRelations. 
i Identification (10) ~ObjectAttribute (9) ElabObject ObjectFunction (3) ~SetMember (3) ElabPart~ProcessStep (5) ~WholePart (8) 
\ElabGenerality~GeneralSpecific (14) 
AbstractInstance (13) 
Interpretation (2)--Evaluation (3) 
S-mmary (4) 
Restatement (9)/Conclusion (7) 
_Location (6) 
. ~Time (7) 
Circumstance (4)E=---Means (4) 
/ ~ Manner (2) 
/ ~Instrument (1) 
Enhancement ~Background (4) (3) (1)~ ParallelEvent 
~Solutionhood (1),,, Answer (1) 
SeqTemporal (6) 
Sequence (6)<SeqSpatial (I) 
r SeqOrdinal (3) 
VolCause (I) 
C/RVol~VolResult (I) 
Cause/Result (14)~ NonVolCause (I) 
C/RN°nV°l~NonVolResult(2 ) 
Purpose (6) 
Enablement (9) 
ension (I', 
/Condition (7) 
GeneralCondition Exception (2) 
Equative (5)   
Contrast (14)~,Antithesis (6) 
Comparative Otherwise (7) 
Comparison (3) , Analogy (4) 
Evidence (7) Proof (I) 
Support (2<Justification (3) 
Motivation (5) 
Concession (6) 
Qualification (1) 
/Conjunction (5) 
gicalRelation I --Disjunction (2) 
131 
by their initials and listed subsequently). In the 
parenthesized comments, S stands for speaker and 
H for hearer. The particular relations defined by 
each researcher and their respective classifications 
are provided in the full version of this paper. 
Elaboration: MT, JH, JG, MP, GH, BF, KD, DSN, QG, 
MH 
Identification: KM, JG, HS, MP, KD, AC, MM, QG, ST, 
RJ 
ElabObject: 
ObjectAttribute: MT, HI, HL, KM, LP, JG, MP, MM, MH 
ObjectFunction: HL, KM, MP 
ElabPart: 
SetMember: MT, KM, JG 
ProcessStep: MT, HP, HI, MP, DL 
WholePart: MT, HI, HL, KM, JG, MP, AC, DL 
ElabGenerality: 
GeneralSpecific: MT, HP, JH, KM, JG, TNR, HS, MP, KD, 
AC, NS, RC, QG, MH 
AbstractInstance: MT, HP, JH, KM, LP, TNR, JG, HS, 
MP, MM, RC, QG, MH 
Interpretation: MT, KD 
Evaluation (S opinion): MT, KD, JH 
Restatement: MT, KM, KD, DSN, NS, RR, RC, QG, MH 
Summary (short restatement): MT, DSN, RC, QG 
Conclusion (interp at end): KM, JG, HS, KD, RR, RC, QG 
Enhancement: MH 
Circumstance: MT, JG, DSN, QG 
Location: HI, HL, KD, QG, RJ, MH 
Time: HI, HL, TNR, KD, QG, RJ, MH 
Means: MP, QG, ST, MH 
Manner: QG, MH 
Instrument: QG 
ParalleIEvent: KD, QG, RJ 
Background: MT, JH, HL, MP 
Sohitionhood (general prob): MT 
Answer (numeric prob): KM 
Extension: MH 
Sequence: MT, JH, LP, KD, DSN, RC 
SeqTemporah HI, HP, LP, DL, NS, MH 
SeqSpatiah NS 
SeqOrdinah LP, DSN, QG 
Cause/Result: JH, KM, TNR, JG, GH, KD, LP, RL, RR, 
RC, QG, R J, SA, MH 
C/RVol (volitional): 
VolCause: MT 
VolResult: MT 
C/RNonvol (nonvolitionM): 
NonVolCause: MT 
NonVolResult: MT, MP 
Purpose: MT, HP, KD, QG, SA, MH 
Enablement: MT, JH, HL, TNR, MP, KD, DSN, DL, SA 
GeneralCondition: 
Condition: MT, JG, LP, RL, DL, RC, MH 
Exception: RL, MH 
Comparative: 
Equative (like, while): JG, TNR, DL, QG, MH 
Contrast: MT, JH, LP, IR, TNR, MP, RL, GH, BF, KD, 
NS, DSN, RC, QG 
Antithesis: MT, DSN, JG, HS, KM, QG 
Otherwise (if then else): MT, LP, NS, RL, RC, QG, MH 
Comparison: KM, HS, MH 
Analogy: KM, JG, MP, RR 
Exhortation: 
Support: RR, RC 
Evidence (support claim): MT, KM, JG, MP, BF, KD, ST 
Proof: MP 
Justification (for S act): MT, IR, DL 
Motivation (for H act): MT, MP, DSN, DL, MM 
Concession: MT, DSN, KD, RR, QG, MH 
Qualification: ST 
LogicalRelation: 
Conjunction (Join, and): MT, DSN, RC, QG, MH 
Disjunction: QG, MH 
(Note: Some of the relations of QG and RJ are intra- 
clausal.) 
The above initials refer to the following authors: 
AC: Alison Cawsey \[Cawsey 89\] 
BF: Barbara Fox \[Fox 84\] 
DL: Diane Litman \[Litman 85\] 
DSN: Donia Scott et ai. \[De Souza et al. 89\] 
GH: Graeme Hirst \[Hirst 81\] 
HI: Eduard Hovy, II domain \[Hovy 88c\] 
HL: Eduard Hovy, LILOG domain \[Hovy 89\] 
HP: Eduard Hovy, PEA domain \[Hovy 88a\] 
HS: Shepherd \[Shepherd 26\] 
IR: Ivan Rankin \[Ranldn 89\] 
JG: Joseph Grimes \[Grimes 75\] 
JH: Jerry Hobbs \[Hobbs 78, Hobbs 79, Hobbs 82\] 
KD: Kathleen Dahlgren \[Dahlgren 88\], pp. 178-184 
KM: Kathleen McKeown \[McKeown 85\] 
LP: Livia Polanyi \[Polanyi 88\] 
MH: Michael Halliday \[Halliday 85\], chapter 7 
MM: Mark Maybttry \[Maybury 89\] 
MP: Johanna Moore and Cdcile Paris \[personal communi- 
cation, 1989\], 
\[Moore & Swartout 88, Paris 88, Moore & Paris 89\] 
MT: Bill Mann and Sand_ra Thompson 
\[Mann & Thompson 86, Mann & Thompson 88\] 
NS: Nathalie Simonin \[Simonin 88\] 
QG: Quirk and Greenbaum, pp. 284-296 (mainly) 
\[Quirk & Greenbaum 73\] 
RC: Robin Cohen \[Cohen 83\], appendix II 
R J: Ray Jackendoff \[Jackendoff83\], pp. 166-202 
RL: Robert Longacre \[Longacre 76\] 
RR: Rachel Reichman \[Reiehman 78\], chs. 2,3 
SA: Roger Schank and Robert Abelson, pp. 30-32 
\[Schank & Abelson 77\] 
ST: Stephen Touhnin \[Toulmin 58\], pp. 94-113 
TNR: Sergei Nirenburg et al. \[Tucker et al. 86\] 
Some Evidence for the Hierarchy 
Structure 
Some nonconclusive evidence supports parts of the 
hierarchy, though further study must be done to 
examine all the relations. This evidence is based 
132 
on sensitivity to generalization evinced by many 
cue words and phrases. For example, the cue word 
"then" is associated with SEQUENCE, and can be 
used appropriately to indicate its subordinates SE- 
QTEMPORAL and SEQSPATIAL, as in: 
SEQTEMPORAL: First you play the long 
note, then the short ones 
SEQSPATIAL: On the wall I have a red 
picture, then a blue one 
In contrast, the cue words for the two subrelations 
are specific and cannot be interchanged without in- 
troducing the associated connotation: 
SEQTEMPORAL: After the 10ng note you 
play the short ones 
SEQSPATIAL: Beside the red picture is the 
blue one 
Thus the relation associated with "then" subsumes 
the relations associated with "after" and "beside". 
Similar observations hold for a number of the rela- 
tions (e.g., SOLUTIONHOOD and RESTATEMENT). 
Preliminary investigation indicates possible ad- 
ditional evidence in the syntactic realization of 
some relations: When a relation typically gives rise 
to a dependent clause, then its subrelations tend to 
do so as well. More study must be done by a trained 
linguist. 
Role of Intentional Relations 
What then of the two relations DOMINATES and 
SATISFACTIONPRECEDES? They do not appear 
anywhere in the hierarchy in Figure 2. 
The answer is that these two relations express 
information that is independent of the rhetori- 
cal/semantic meanings of the relations in the tax- 
onomy and only apply in discourses with inten- 
tional, plan-like nature. They derive from early 
work on a highly plan-oriented domain \[Grosz 81\], 
in which plan steps' preconditions led to underly- 
ing precedence orderings of plan steps and satis- 
faction of subgoals which were dominated by su- 
pergoals. However, not all discourse is plan-like; 
a large proportion of everyday close discourse be- 
tween people achieves goals for which, it can be 
argued, no plans can be formulated (for some such 
goals see \[Hovy 88b\]): the banter between friends 
which serves to strengthen interpersonal bonds, the 
discussions in supermarket lines, the language and 
presentation styles employed in order to be friendly 
or attractive, etc. Such communications also ex- 
hibit internal structure, and also employ the rhetor- 
ical/semantic interclausal relations. 
However, it is not clear how to generalize DOMI- 
NATES and SATISFACTIONPRECEDES to cover such 
cases as well. One possible generalization is to use 
the general relations HYPOTAXIS and PARATAXIS 
(that is, asymmetrical and symmetrical relation- 
ships: 
HYPO: Joe left because Sue worked (5 
Sue worked because Joe left) 
PARA: He shouted when the horse 
jumped (= the horse jumped when he 
shouted) 
respectively). But this does not work because both 
DOMINATES and SATISFACTIONPRECEDES are by 
their natures asymmetrical. Another possible gen- 
eralization is to use the two syntactic relations 
MULTIVARIATE and UNIVARIATE (that is, contain- 
ing an embedded syntactic type recursion (a gram- 
matical rank shift) ornot, as in: 
MULTI: Sue's working caused \[Joe to 
leave\] (S embedded in S) 
UNI: Joe left because Sue worked (two 
coequal Ss) 
respectively). This does not work either because 
some DOMINATES relations hold between syntacti- 
cally independent complete sentences, so no syn- 
tactic embedding occurs. 
Does this mean that the two intentional relations 
should simply be added into the hierarchy? No, 
because they can be realized in the discourse by 
various rhetorical/semantic relations; for example, 
SATISFACTIONPREcEDEs can be expressed by SE- 
QUENCE, as in: 
First boil the water, then add the rice, and 
then stir 
or by CAUSE, as in: 
The sun heats the desert, which causes the 
air to rise, which causes a breeze 
Thus, in the absence of other candidate gener- 
alizations, one can conclude that the relations 
DOMINATES and SATISFACTIONPRECEDES are in- 
dependent of the rhetorical/semantic taxonomiza- 
tion and provide an additional dimension of infor- 
mation, used for those discourses that exhibit an 
appropriately intentional plan-like nature. 
Some Objections Answered 
A number of objections may be raised to the tax- 
onomization presented here. I attempt to respond 
to some of the more serious ones: 
133 
Objection 1. Does it make sense at all to or- 
ganize into a single taxonomy such disparate en- 
tities? After all, RESTATEMENT and CONCESSION 
are primarily rhetorical while, say, PURPOSE and 
PROCESSSTEP are primarily semantic. 
Why not? As a result of this study and previ- 
ous work, I believe that "rhetorical" relations are 
simply the crudest or most generalized "semantic" 
ones; in other words, that no purely rhetorical or 
purely semantic relations exist. Some relations, 
typically those higher in the hierarchy, certainly 
fulfill a more rhetorical function (i.e., provide more 
information about the argumentational structure of 
the discourse) than a semantic function. But not 
even the topmost relations are entirely devoid of se- 
mantic meaning. Since all the relations have some 
semantic import (regardless of their structural im- 
port), this objection to their being arranged into a 
single taxonomy for the purposes of discourse struc- 
ture description is groundless 1. 
Objection 2. What guarantee exists that the 
relations given in the taxonomy are indeed the 
"right" ones? Or the only ones? It is not diffi- 
cult to come up with relations that differ in some 
way from those in the taxonomy and that do not 
neatly fall under a single item in it. 
This is a standard objection to any set of terms 
proposed to fulfill some function. The standard re- 
sponse holds here too: there is no guarantee that 
these are the "right" relations, whatever "right" 
may mean. Similarly, there is no guarantee that 
the terms \[VERB NOUN ADJECTIVE ADVERB ... \] 
are the "right" and "only" terms for types of words; 
they have simply been canonized by long use and 
much experience. There is enough evidence from 
actual attempts at constructing working systems 
(text planners and discourse analyzers) that rela- 
tions at this level of interclausal representation are 
required to guide inference and planning processes. 
Without such relations we simply cannot construct 
an adequate account of the structure of a discourse 
1This position is in contradistinction with that of Sys- 
temic Linguistics, which holds that semantic relations are of 
the type Ideational while rhetorical relations are of the type 
Textual. Since both kinds of relations serve the same func- 
tion, namely to bind together different pieces of knowledge, 
and certain relations operate both interclausally and intra- 
clausally, there seems to me no reason to differentiate them 
into two different types. I categorize discourse structure re- 
lations with those called Ideational, and group separately 
all those phenomena that reflect the characteristics of the 
communicative setting: referentially available artifacts and 
concepts, the nature of the medium (telephone, paper, con- 
versation, etc.), and the physical and social context (back- 
ground noise, available time, etc.). 
nor plan an adequate multisentence paragraph by 
computer. 
The particular relations proposed here are cer- 
tainly open to question, and their strongest support 
is that they are the amalgamation and synthesis of 
the efforts and proposed terms of over 25 different 
investigators from different fields, as noted previ- 
ously. In addition, there is always the possibility 
that new interclausal relations will be needed that 
cannot in fact be subsumed under existing nodes 
in the taxonomy. While not impossible, I believe 
this is unlikely, based on my experience in compil- 
ing the hierarchy: After the top three levels had 
more or less been established halfway through this 
study, only one new second-level relation -- IDEN- 
TIFICATION -- had to be added. I expect that when 
new domains are investigated, the hierarchy will 
grow primarily at the bottom, and that the ratio 
of the number of relations added at one level to the 
number of relations added at the next lower level, 
averaged across all levels, will be well below 0.2. 
Objection 3. The taxonomy is unbounded to- 
ward the bottom: it places one on the slippery 
slope toward having to deal with the full complex- 
ity of semantic meaning. Simply working on the 
structure of discourse is difficult enough without 
bringing in the complexity of semantic knowledge. 
This is the the Parsimonious Position objection. 
There is no reason to fear the complexity of an 
unbounded set of terms, whether semantic or not, 
as long as the terms are well-behaved and sub- 
ject to a pattern of organization which makes them 
manageable. A hierarchicalization of the terms in 
which all the pertinent information about discour- 
sal behavior is captured at the top (which is maxi- 
mally general, bounded, and well-understood) and 
not at the bottom (which permits unboundedness 
and redundancy) presents no threat to computa- 
tional processing. Each discourse relation simply 
inherits from its ancestors all necessary process- 
ing information, such as cue words and realization 
constraints, adding its unique peculiarities, to be 
used for inferring its discoursal role (in parsing) or 
for planning out a discourse (in generation). In- 
creasing differentiation of relations, continued un- 
til the very finest nuances of meaning are sepa- 
rately represented, need be pursued only to the 
extent required for any given application. Thus 
"unbounded" growth of semantic relations is not a 
problem, as long as they can be subsumed under 
existing nodes in the taxonomy. 
Objection 4. The hierarchy contains no explic- 
itly structural relations, yet explicit signals of dis- 
134 
course structure are very common. Why are they 
not included? 
In fact, the relations people use to signal dis- 
course structure (such as parallelism, say) are in- 
cluded in the taxonomy. The most popular relation 
(one that appeared as a separate relation in a num- 
ber of the studies cited) is PARALLEL, which is in- 
variably signalled using some set of SEQUENCE rela- 
tions, often after having been explicitly introduced 
by a separate clause. The fact that SEQUENCE 
relations can be used to signal both semantic se- 
quentiality and rhetorical structure is no reason for 
alarm; if conditions should arise in which the two 
functions require different behavior, a more specific 
relation subordinated to SEQUENCE, dedicated to 
expressing purely rhetorical sequentiality (such as, 
say, SEQSTRUCTURAL), can be created, though it 
is likely that in the actual presentation, the rhetor- 
ical sequence will usually follow some temporal or 
spatial sequence in any case. 
Conclusion 
A rather gratifying result of the synthesis presented 
here is that only 16 core relations, organized into 
3 principal types, suffice to cover essentially all the 
interclausal relations proposed by the sources. This 
suggests that other relations not yet in the hierar- 
chy are likely to be subtypes of relations already 
in it, preserving the boundedness of the number of 
relation types. The relations are rhetorical in na- 
ture, becoming increasingly semantic as they be- 
come more specific; the claim is that the rhetorical 
relations are simply least delicate (i.e., most gen- 
eral) semantic relations. 
While some evidence is provided for the struc- 
ture of the hierarchy, as well as for the claim that 
the relations are independent of the goal-oriented 
plan-based discourse structure relations proposed 
by \[Grosz & Sidner 86\], there is no claim that this 
hierarchy is complete or correct in all details. It 
is certainly open to elaboration, enhancement, and 
extension! The hope is that it will serve the com- 
munity by proving to be a common starting point 
and straw man for future work on discourse struc- 
ture. 
Acknowledgments 
Thanks to Christian Matthiessen, John Bateman, 
Robin Cohen, Kathleen McCoy, Kathleen McKe- 
own, Johanna Moore, Mick O'Donnell, and C4cile 
Paris, and to everyone who sent me their relations. 
I am still collecting relations and will continue to 
update this hierarchy. 

References 
\[Aristotle 54\] Aristotle. The Rhetoric. In The Rhetoric 
and the Poetics of Aristotle, W. Rhys Roberts 
(trans), Random House, New York, 1954. 
\[Cawsey 89\] Cawsey, A. Generating Communicative 
Discourse. Presented at the Second European Work- 
shop on Language Generation, Edinburgh, 1989. In 
Mellish, C., Dale, R., and Zock, M. (eds), selected 
papers from the workshop, in prep. 
\[Cohen 83\] Cohen, R. A Computational Model for the 
Analysis of Arguments. Technical Report CSRG-151, 
University of Toronto, Toronto, 1983. 
\[Dahlgren 88\] Dahlgren, K. Naive Semantics for Nat- 
ural Language Understanding. Kluwer Academic 
Press, Boston, 1988. 
\[De Souza et al. 89\] De Souza, C.S., Scott, D.R. 
and Nunes, M.G.V. Enhancing Text Quality 
in a Question-Answering System. Unpublished 
manuscript, Pontificia Universidade Cat61ica de Rio 
de Janeiro, 1989. 
\[Fox 84\] Fox, B. Discourse Structure and Anaphora in 
Written and Conversational English. Ph.D. disserta- 
tion, UCLA, Los Angeles, 1984. 
\[Grimes 75\] Grimes, J.E. The Thread of Discourse. 
Mouton, The Hague, 1975. 
\[Grosz 81\] Grosz, B.J. Focusing and Description in 
Natural Language Dialogues. In Joshi, A., Webber, 
B., and Sag, I. (eds) Elements of Discourse Under- 
standing. Cambridge University Press, Cambridge, 
1981. 
\[Grosz & Sidner 86\] Grosz, B.J. and Sidner, C.L. At- 
tention, Intentions, and the Structure of Discourse. 
In Journal o\] Computational Linguistics 12(3), 1986 
(175-204). 
\[Halliday 85\] Halliday, M.A.K. An Introduction to 
Functional Grammar. Edward Arnold Press, Balti- 
more, 1985. 
\[Hirst 81\] Hirst, G. Discourse-Oriented Anaphora Res- 
olution: A Review. In Journal of Computational Lin- 
guistics 7, 1981 (85-98). 
\[Hobbs 78\] Hobbs, J.R. Why is Discourse Coherent? 
Technical Note no. 176, SRI International, Menlo 
Park CA, 1978. 
\[Hobbs 79\] Hobbs, J.R. Coherence and Coreference. 
Cognitive Science 3(1), 1979 (67-90). 
\[Hobbs 82\] Hobbs, J.R. Coherence in Discourse. In 
Strategies for Natural Language Processing, Lehnert, 
W.G. and Ringle, M.H. (eds), Lawrence Erlbaum 
Associates, Hillsdale, 1982 (223-243). 
\[Hovy 88a\] Hovy, E.H. Planning Coherent Multisen- 
tential Text. In Proceedings of the 26th ACL Con- 
ference, Buffalo, 1988 (163-169). 
\[Hovy 88b\] Hovy, E.H. Two Types of Planning in Lan- 
guage Generation. In Proceedings of the 26th A CL 
Conference, Buffalo, 1988 (170-176). 
\[Hovy 88c\] Hovy, E.H. Approaches to the Planning of 
Coherent Text. Presented at the $th International 
Workshop on Text Generation, Los Angeles, 1988. In 
Paris, C.L., Swartout, W.R. and Mann, W.C. (eds), 
Natural Language in Artificial Intelligence and Com- 
putational Linguistics, to appear. 
\[Hovy 89\] Hovy, E.H. Notes on Dialogue Management 
and Text Planning in the LILOG Project. Unpub- 
lished working document, Projekt LILOG, Insti- 
tut fiir Wissensbasierte Systeme, IBM Deutschland, 
Stuttgart, 1989. 
\[Jackendoff 83\] Jackendoff, R. Semantics and Cogni- 
tion. MIT Press, Cambridge, 1983. 
\[Litman 85\] Litman, D. Plan Recognition and Dis- 
course Analysis: An Integrated Approach for Under- 
standing Dialogues. Ph.D. dissertation, University of 
Rochester, Rochester, 1985. 
\[Longacre 76\] Longacre, R. An Anatomy of Speech No- 
tions. Peter de Ridder Press, Lisse, 1976. 
\[Mann & Thompson 86\] Mann, W.C. & Thompson, 
S.A. Rhetorical Structure Theory: Description and 
Construction of Text Structures. In Natural Lan- 
guage Generation: New Results in Artificial In- 
telligence, Psychology, and Linguistics, Kempen, 
G. (ed), Kluwer Academic Publishers, Dordrecht, 
Boston, 1986 (279-300). 
\[Mann & Thompson 88\] Mann, W.C. and Thompson, 
S.A. Rhetorical Structure Theory: Toward a Func- 
tional Theory of Text Organization. Text 8(3), 1988 
(243-281). Also available as USC/Information Sci- 
ences Institute Research Report RR-87-190. 
\[Maybury 89\] Maybury, M.T. Enhancing Explanation 
Coherence with Rhetorical Strategies. In Proceedings 
of the European ACL Conference, Manchester, 1989 
(168-173). 
\[McKeown 85\] McKeown, K.R. Text Generation: Us- 
ing Discourse Strategies and Focus Constraints to 
Generate Natural Language Text. Cambridge Univer- 
sity Press, Cambridge, 1985. 
\[Moore & Paris 89\] Moore, J.D. and Paris, C.L. Plan- 
ning Text for Advisory Dialogues. In Proceedings of 
the $7th ACL Conference, Vancouver, 1989. 
\[Moore & Swartout 88\] Moore, J.D. and Swartout, 
W.R. Dialogue-Based Explanation. Presented at the 
4th International Workshop on Text Generation, Los 
Angeles, 1988. In Paris, C.L., Swartout, W.R. and 
Mann, W.C. (eds), Natural Language in Artificial In- 
telligence and Computational Linguistics, to appear. 
\[Paris 88\] Paris, C.L. Generation and Explanation: 
Building an Explanation Facility for the Explain- 
able Expert Systems Framework. Presented at the 
4th International Workshop on Text Generation, Los 
Angeles, 1988. In Paris, C.L., Swartout, W.R. and 
Mann, W.C. (eds), Natural Language in Artificial In- 
telligence and Computational Linguistics, to appear. 
\[Polanyi 88\] Polanyi, L. A formal Model of the Struc- 
ture of Discourse. Journal of Pragmatics 12, 1988 
(601-638). 
\[Quirk & Greenbaum 73\] Quirk, R., and Greenbaum, 
S. A Concise Grammar of Contemporary English. 
Harcourt Brace Jovanovich Inc., New York, 1973. 
\[Rankin 89\] Rankin, I. The Deep Generation of Text 
in Expert Critiquing Systems. Licentiate thesis, Uni- 
versity of Link6ping, Sweden, 1989. 
\[Reichman 78\] 
Reichman, R. Conversational Coherency. Cognitive 
Science 2, 1978 (283-327). 
\[Schank & Abelson 77\] Schank, R.C. and Abelson, R. 
Scripts, Plans, Goals, and Understanding. Lawrence 
Erlbaum Associates, Hillsdale, 1977. 
\[Shepherd 26\] Shepherd, H.R. The Fine Art of Writ- 
ing. The Macmillan Co., New York, 1926. 
\[Simonin 88\] Simonin, N. An Approach for Creating 
Structured Text. In Zock, M. and Sabah, G. (eds), 
Advances in Natural Language Generation vol. 1, 
Pinter Publishers, London, 1988 (146-160). 
\[Toulmin 58\] Toulmin, S. The Uses of Argument. Cam- 
bridge University Press, Cambridge, 1958. 
\[Tucker et al. 86\] Tucker, A.B., Nirenburg, S., and 
Raskin, V. Discourse and Cohesion in Expository 
Text. In Proceedings of Coling-86, 1986 (181-183). 
